@kk-irving/knowledge-mcp-server 1.0.0

package/dist/analyze-issue.js

@@ -0,0 +1,392 @@
+/**
+ * analyze_issue —— 端到端 PR/Bug 分析编排（v1.0.0）。
+ *
+ * 一次调用串起：
+ *   1. 拉 Zmind issue 详情（subject + description + journals + attachments）
+ *   2. 准备工作目录（默认 `<workspace_root>/.workspace/issue-<id>/`）
+ *   3. 提取查询关键词（标题去停用词 + 描述前 200 字符 token 化）
+ *   4. 三源本地知识库混合检索（zmind / gerrit / confluence）
+ *   5. 推断潜在 module（从命中的 gerrit/zmind 路径片段反查 module-map）
+ *   6. （可选）AOSP 模块级精搜
+ *   7. 渲染 `analysis-context.md` 落盘到工作目录
+ *   8. 返回 JSON 汇总
+ *
+ * 任何子步骤失败均 best-effort 继续，错误写入 context.md "已知问题" 段。
+ *
+ * 注意：本工具**不下载附件**——附件下载由 zmind-mcp-server 的
+ * `prepare_issue_workspace` 负责，此处只读取已有 workspace。如果调用时
+ * workspace_root 不存在则只创建目录。
+ */
+import { mkdir, writeFile, readdir, stat } from "node:fs/promises";
+import * as path from "node:path";
+import { searchLocal } from "./search.js";
+import { searchAosp } from "./aosp/search.js";
+import { loadModuleMap } from "./aosp/module-map-loader.js";
+const ZMIND_URL = (process.env.ZMIND_URL ?? "https://zmind.whaletv.com").replace(/\/+$/, "");
+const ZMIND_API_KEY = process.env.ZMIND_API_KEY ?? "";
+// =============================================================================
+// Zmind helper
+// =============================================================================
+async function fetchIssue(issueId) {
+    if (!ZMIND_API_KEY)
+        throw new Error("ZMIND_API_KEY 未配置");
+    const url = new URL(`/issues/${issueId}.json`, ZMIND_URL + "/");
+    url.searchParams.set("include", "attachments,journals");
+    url.searchParams.set("key", ZMIND_API_KEY);
+    const res = await fetch(url.toString(), {
+        headers: { Accept: "application/json" },
+    });
+    if (!res.ok) {
+        throw new Error(`Zmind HTTP ${res.status}: ${(await res.text()).slice(0, 300)}`);
+    }
+    const data = (await res.json());
+    return data.issue;
+}
+// =============================================================================
+// 关键词提取
+// =============================================================================
+const STOPWORDS_ZH = new Set([
+    "的", "了", "和", "是", "在", "有", "也", "就", "都", "这", "那", "我", "你", "他",
+    "不", "为", "上", "下", "之", "对", "以", "及", "及其", "或", "等", "与",
+    "请", "问题", "情况", "发现", "出现", "存在", "需要", "处理", "解决", "测试",
+    "已", "未", "可", "能", "会", "时", "时候",
+]);
+const STOPWORDS_EN = new Set([
+    "the", "a", "an", "is", "are", "was", "were", "be", "been", "being",
+    "of", "to", "in", "on", "at", "for", "with", "and", "or", "but",
+    "this", "that", "these", "those", "it", "its", "as",
+    "i", "we", "you", "they", "he", "she",
+    "have", "has", "had", "do", "does", "did", "can", "will", "would", "should",
+    "issue", "bug", "test", "fix", "fail", "failed", "problem",
+]);
+/**
+ * 从 issue 标题 + 描述前 200 字符提取关键词。
+ *
+ * 简单策略：
+ *   - 把内容按非字母数字字符切分
+ *   - 同时切分中文（按字符为单位，但保留连续中文片段）
+ *   - 过滤停用词、长度 < 2 的 token
+ *   - 取前 8 个去重 token
+ */
+export function extractKeywords(subject, description) {
+    const text = (subject + " " + (description || "").slice(0, 200)).toLowerCase();
+    // 同时切英文 token 与连续中文片段
+    const englishTokens = (text.match(/[a-z][a-z0-9_-]+/g) ?? []).filter((t) => t.length >= 2 && !STOPWORDS_EN.has(t));
+    const chineseTokens = (text.match(/[\u4e00-\u9fff]+/g) ?? []).filter((t) => t.length >= 2 && !STOPWORDS_ZH.has(t));
+    const seen = new Set();
+    const result = [];
+    for (const t of [...englishTokens, ...chineseTokens]) {
+        if (seen.has(t))
+            continue;
+        seen.add(t);
+        result.push(t);
+        if (result.length >= 8)
+            break;
+    }
+    return result;
+}
+// =============================================================================
+// 平台 / 模块推断
+// =============================================================================
+const PLATFORM_KEYWORDS = {
+    D4: ["d4", "am30", "at30", "calla", "redi", "soddy", "t982"],
+    X5: ["x5", "am50", "br30", "bs30", "anemone", "dahlia", "daisy", "dryas"],
+    STB: ["stb", "stb16", "pascal", "qurra", "raman", "ross"],
+};
+function inferPlatform(text) {
+    const lower = text.toLowerCase();
+    for (const [platform, keywords] of Object.entries(PLATFORM_KEYWORDS)) {
+        for (const kw of keywords) {
+            if (lower.includes(kw))
+                return platform;
+        }
+    }
+    return undefined;
+}
+/**
+ * 从命中的搜索结果（含 gerrit project / 文件路径片段）反查可能的 module。
+ */
+async function inferModulesFromHits(hits, platform) {
+    if (!platform || hits.length === 0)
+        return [];
+    let map;
+    try {
+        map = await loadModuleMap();
+    }
+    catch {
+        return [];
+    }
+    const platformMap = map.platforms[platform.toUpperCase()];
+    if (!platformMap)
+        return [];
+    const moduleScores = new Map();
+    // 对每个命中结果，扫描其 project / 路径片段，看是否匹配 module path 前缀
+    for (const hit of hits) {
+        const text = String((hit?.project ?? "") + " " + (hit?.snippet ?? "") + " " + (hit?.title ?? "")).toLowerCase();
+        for (const [moduleName, paths] of Object.entries(platformMap)) {
+            let hitCount = 0;
+            for (const p of paths) {
+                const pLower = p.toLowerCase();
+                // 取 path 最后一段作为最特征性 token
+                const segs = pLower.split("/").filter(Boolean);
+                for (const seg of segs.slice(-2)) {
+                    if (seg.length >= 4 && text.includes(seg)) {
+                        hitCount++;
+                        break;
+                    }
+                }
+                if (text.includes(moduleName))
+                    hitCount++;
+            }
+            if (hitCount > 0) {
+                moduleScores.set(moduleName, (moduleScores.get(moduleName) ?? 0) + hitCount);
+            }
+        }
+    }
+    return Array.from(moduleScores.entries())
+        .sort((a, b) => b[1] - a[1])
+        .slice(0, 3)
+        .map(([m]) => m);
+}
+// =============================================================================
+// 工作目录
+// =============================================================================
+async function ensureWorkspace(workspaceRoot, issueId) {
+    const wsRoot = path.resolve(workspaceRoot);
+    const target = path.join(wsRoot, ".workspace", `issue-${issueId}`);
+    await mkdir(target, { recursive: true });
+    return target;
+}
+async function summarizeAttachments(workspacePath) {
+    const attachDir = path.join(workspacePath, "attachments");
+    try {
+        const st = await stat(attachDir);
+        if (!st.isDirectory())
+            return { count: 0, sample: [] };
+    }
+    catch {
+        return { count: 0, sample: [] };
+    }
+    const entries = await readdir(attachDir, { withFileTypes: true });
+    const files = entries.filter((e) => e.isFile()).map((e) => e.name);
+    return { count: files.length, sample: files.slice(0, 5) };
+}
+// =============================================================================
+// Markdown 渲染
+// =============================================================================
+function renderContextMd(args) {
+    const lines = [];
+    const issue = args.issue;
+    lines.push(`# Issue #${issue.id} 分析上下文`);
+    lines.push("");
+    lines.push(`- **标题**: ${issue.subject ?? "(无)"}`);
+    lines.push(`- **类型**: ${issue.tracker?.name ?? ""}`);
+    lines.push(`- **状态**: ${issue.status?.name ?? ""}`);
+    lines.push(`- **指派**: ${issue.assigned_to?.name ?? ""}`);
+    lines.push(`- **项目**: ${issue.project?.name ?? ""}`);
+    lines.push(`- **目标版本**: ${issue.fixed_version?.name ?? ""}`);
+    lines.push(`- **生成时间**: ${new Date().toISOString()}`);
+    lines.push(`- **工作目录**: ${args.workspace}`);
+    lines.push("");
+    // 描述
+    lines.push("## 问题描述");
+    lines.push("");
+    lines.push(issue.description ?? "(无描述)");
+    lines.push("");
+    // 附件
+    lines.push("## 附件");
+    lines.push("");
+    if (args.attachments.count === 0) {
+        lines.push("（暂无）— 如需下载附件请用 zmind-mcp 的 `prepare_issue_workspace` 工具");
+    }
+    else {
+        lines.push(`共 ${args.attachments.count} 个；样本：`);
+        for (const f of args.attachments.sample)
+            lines.push(`- \`${f}\``);
+    }
+    lines.push("");
+    // 关键词与平台/模块
+    lines.push("## 推断信息");
+    lines.push("");
+    lines.push(`- **关键词**: ${args.keywords.length > 0 ? args.keywords.join(", ") : "(空)"}`);
+    lines.push(`- **平台**: ${args.platform ?? "(未推断)"}`);
+    lines.push(`- **可能模块**: ${args.modules.length > 0 ? args.modules.join(", ") : "(未推断)"}`);
+    lines.push("");
+    // 相似历史
+    lines.push("## 相似历史（本地知识库 hybrid 检索）");
+    lines.push("");
+    for (const sourceName of ["zmind", "gerrit", "confluence"]) {
+        const hits = args.similar[sourceName] ?? [];
+        lines.push(`### ${sourceName} (${hits.length} 条)`);
+        lines.push("");
+        if (hits.length === 0) {
+            lines.push("（无命中）");
+        }
+        else {
+            for (const h of hits) {
+                const titlePart = h.title ?? "(无标题)";
+                const url = h.url ?? "";
+                const snippet = (h.snippet ?? "").replace(/\n/g, " ");
+                const score = typeof h.score === "number" ? h.score.toFixed(3) : "";
+                lines.push(`- **[${h.id}]** [${titlePart}](${url}) · score=${score} · match=${h.match}`);
+                if (snippet)
+                    lines.push(`  - ${snippet.slice(0, 240)}`);
+            }
+        }
+        lines.push("");
+    }
+    // AOSP
+    lines.push("## AOSP 代码（模块级精搜）");
+    lines.push("");
+    if (args.aospHits.length === 0) {
+        lines.push("（未启用 / 未命中）");
+    }
+    else {
+        for (const h of args.aospHits) {
+            lines.push(`- **${h.module}** ${h.file_path}:${h.line_start}-${h.line_end} (${h.symbol_kind} ${h.symbol_name})`);
+            const snip = (h.snippet ?? "").replace(/\n/g, " ");
+            if (snip)
+                lines.push(`  - ${snip.slice(0, 240)}`);
+        }
+    }
+    lines.push("");
+    // 后续建议
+    lines.push("## 推荐动作");
+    lines.push("");
+    lines.push("1. **复现验证** — 在 issue 描述场景下尝试本地复现");
+    lines.push("2. **历史修复对比** — 重点查看上面三源 Top-K 命中里 status=Closed 的条目，对比修复方式");
+    lines.push("3. **模块改动建议** — 结合 AOSP 命中代码片段定位修改点；如未命中可手动用 `search_aosp` 缩小范围");
+    lines.push("");
+    // 已知问题
+    if (args.errors.length > 0) {
+        lines.push("## 已知问题（运行期错误）");
+        lines.push("");
+        for (const e of args.errors) {
+            lines.push(`- **${e.stage}**: ${e.message}`);
+        }
+        lines.push("");
+    }
+    return lines.join("\n");
+}
+// =============================================================================
+// 主入口
+// =============================================================================
+export async function analyzeIssue(args) {
+    const errors = [];
+    const issueId = args.issue_id;
+    const workspaceRoot = args.workspace_root ?? process.cwd();
+    const perSourceLimit = Math.max(1, Math.min(10, args.per_source_limit ?? 3));
+    // 1. 拉 issue
+    let issue;
+    try {
+        issue = await fetchIssue(issueId);
+    }
+    catch (e) {
+        errors.push({ stage: "fetch_issue", message: e.message });
+        issue = { id: issueId };
+    }
+    // 2. 工作目录
+    const wsPath = await ensureWorkspace(workspaceRoot, issueId);
+    // 3. 附件汇总
+    let attachSummary = { count: 0, sample: [] };
+    try {
+        attachSummary = await summarizeAttachments(wsPath);
+    }
+    catch (e) {
+        errors.push({ stage: "summarize_attachments", message: e.message });
+    }
+    // 4. 关键词
+    const keywords = extractKeywords(issue.subject ?? "", issue.description ?? "");
+    // 5. 平台推断
+    const platformText = `${issue.subject ?? ""} ${issue.description ?? ""} ${issue.project?.name ?? ""} ${issue.fixed_version?.name ?? ""}`;
+    const inferredPlatform = (args.platform ?? inferPlatform(platformText) ?? "").toUpperCase() || undefined;
+    // 6. 三源 hybrid 检索
+    let similar = {};
+    if (keywords.length > 0) {
+        try {
+            const query = keywords.join(" ");
+            const result = (await searchLocal({
+                query,
+                source: "all",
+                mode: "hybrid",
+                limit: perSourceLimit,
+            }));
+            similar = {
+                zmind: result.zmind ?? [],
+                gerrit: result.gerrit ?? [],
+                confluence: result.confluence ?? [],
+            };
+        }
+        catch (e) {
+            errors.push({ stage: "search_local", message: e.message });
+        }
+    }
+    // 7. 模块推断
+    const allHitsForInference = [
+        ...(similar.zmind ?? []),
+        ...(similar.gerrit ?? []),
+    ];
+    let modules = [];
+    try {
+        modules = await inferModulesFromHits(allHitsForInference, inferredPlatform);
+    }
+    catch (e) {
+        errors.push({ stage: "infer_modules", message: e.message });
+    }
+    // 8. AOSP 检索（可选）
+    let aospHits = [];
+    if (args.include_aosp && inferredPlatform && modules.length > 0 && keywords.length > 0) {
+        try {
+            const result = await searchAosp({
+                query: keywords.join(" "),
+                platform: inferredPlatform,
+                module: modules[0],
+                mode: "hybrid",
+                limit: perSourceLimit,
+            });
+            aospHits = result.hits;
+        }
+        catch (e) {
+            errors.push({ stage: "search_aosp", message: e.message });
+        }
+    }
+    // 9. 渲染 context.md
+    const md = renderContextMd({
+        issue,
+        workspace: wsPath,
+        attachments: attachSummary,
+        keywords,
+        platform: inferredPlatform,
+        modules,
+        similar,
+        aospHits,
+        errors,
+    });
+    const ctxPath = path.join(wsPath, "analysis-context.md");
+    try {
+        await writeFile(ctxPath, md, "utf8");
+    }
+    catch (e) {
+        errors.push({ stage: "write_context_md", message: e.message });
+    }
+    return {
+        workspace_path: wsPath,
+        issue: {
+            id: issueId,
+            tracker: issue.tracker?.name ?? "",
+            subject: issue.subject ?? "",
+            status: issue.status?.name ?? "",
+            assignee: issue.assigned_to?.name ?? "",
+            project: issue.project?.name ?? "",
+            target_version: issue.fixed_version?.name ?? "",
+        },
+        attachments_summary: attachSummary,
+        keywords,
+        inferred_platform: inferredPlatform,
+        inferred_modules: modules,
+        similar,
+        aosp_hits: aospHits,
+        context_md_path: ctxPath,
+        errors,
+    };
+}

package/dist/aosp/chunker.js

@@ -0,0 +1,221 @@
+/**
+ * 文件切块（chunker）。
+ *
+ * 设计：
+ *   - 不上 tree-sitter（500MB+ 体积、需各语言 wasm）
+ *   - 用 **正则边界识别 + 行级硬切**，足够覆盖 AOSP 主流语言（Java/Kotlin/C/C++/Python/Shell/AIDL/HIDL/XML 注释边界）
+ *   - 每个 chunk ≤ MAX_CHUNK_CHARS（默认 2000 字符）
+ *   - 优先在边界（class / fn / method）切；落不到边界时按 200 行硬切
+ *   - 每个 chunk 记录 file_path、line_start、line_end、symbol_kind、symbol_name
+ *
+ * 黑名单：在 indexer 层处理（跳过 .git / out / build / 二进制 / >5MB）
+ */
+const MAX_CHUNK_CHARS = 2000;
+const HARD_LINES = 200;
+// =============================================================================
+// 边界识别（正则）
+// =============================================================================
+function detectBoundariesJava(lines) {
+    // class / interface / enum / methods
+    const result = [];
+    // 顶层 class / interface / enum
+    const classRe = /^\s*(?:public|private|protected|abstract|final|static|sealed)?\s*(class|interface|enum|@interface)\s+(\w+)/;
+    // method（有 paren 的非 class 行；不严格，但够用）
+    const methodRe = /^\s*(?:public|private|protected|static|final|abstract|synchronized|default|native)\s+(?:[\w<>?,\s.\[\]]+\s+)?(\w+)\s*\(/;
+    for (let i = 0; i < lines.length; i++) {
+        const ln = lines[i];
+        if (!ln)
+            continue;
+        let m = ln.match(classRe);
+        if (m) {
+            result.push({ line: i + 1, kind: m[1], name: m[2] });
+            continue;
+        }
+        m = ln.match(methodRe);
+        if (m && !ln.trimStart().startsWith("//")) {
+            // 排除 control statement
+            if (!/^\s*(if|for|while|switch|catch|synchronized|try)\s*\(/.test(ln)) {
+                result.push({ line: i + 1, kind: "method", name: m[1] });
+            }
+        }
+    }
+    return result;
+}
+function detectBoundariesKotlin(lines) {
+    const result = [];
+    const classRe = /^\s*(?:open|abstract|sealed|data|enum|object)?\s*(class|interface|object|enum)\s+(\w+)/;
+    const funRe = /^\s*(?:override|open|private|public|protected|internal|suspend|inline)?\s*fun\s+(?:<[^>]*>\s*)?(?:(\w+)\.)?(\w+)\s*\(/;
+    for (let i = 0; i < lines.length; i++) {
+        const ln = lines[i];
+        if (!ln)
+            continue;
+        let m = ln.match(classRe);
+        if (m) {
+            result.push({ line: i + 1, kind: m[1], name: m[2] });
+            continue;
+        }
+        m = ln.match(funRe);
+        if (m) {
+            result.push({ line: i + 1, kind: "function", name: m[2] || m[1] || "" });
+        }
+    }
+    return result;
+}
+function detectBoundariesCpp(lines) {
+    const result = [];
+    // class / struct
+    const classRe = /^\s*(?:template\s*<[^>]*>\s*)?(class|struct)\s+(\w+)/;
+    // 函数定义（含返回类型）；非常宽松：以 ) { 结尾的行
+    const funRe = /^\s*(?:[\w:<>*&\s,]+)\s+(\w+)\s*\([^)]*\)\s*(?:const)?\s*(?:override)?\s*\{?\s*$/;
+    for (let i = 0; i < lines.length; i++) {
+        const ln = lines[i];
+        if (!ln)
+            continue;
+        let m = ln.match(classRe);
+        if (m) {
+            result.push({ line: i + 1, kind: m[1], name: m[2] });
+            continue;
+        }
+        m = ln.match(funRe);
+        if (m && !ln.trim().startsWith("//") && !/^\s*(if|for|while|switch|catch|return)\s*\(/.test(ln)) {
+            // 至少要看像函数签名（含返回类型字符）：放到稍后 dedup
+            if (ln.includes(" ") || ln.includes(":")) {
+                result.push({ line: i + 1, kind: "function", name: m[1] });
+            }
+        }
+    }
+    return result;
+}
+function detectBoundariesPython(lines) {
+    const result = [];
+    const re = /^\s*(class|def|async def)\s+(\w+)/;
+    for (let i = 0; i < lines.length; i++) {
+        const ln = lines[i];
+        if (!ln)
+            continue;
+        const m = ln.match(re);
+        if (m) {
+            result.push({ line: i + 1, kind: m[1].includes("class") ? "class" : "function", name: m[2] });
+        }
+    }
+    return result;
+}
+function detectBoundariesGeneric(_lines) {
+    return []; // 未识别语言走纯硬切
+}
+function pickDetector(fileExt) {
+    switch (fileExt.toLowerCase()) {
+        case ".java":
+        case ".aidl":
+        case ".hal":
+            return detectBoundariesJava;
+        case ".kt":
+        case ".kts":
+            return detectBoundariesKotlin;
+        case ".c":
+        case ".cc":
+        case ".cpp":
+        case ".cxx":
+        case ".h":
+        case ".hh":
+        case ".hpp":
+            return detectBoundariesCpp;
+        case ".py":
+            return detectBoundariesPython;
+        default:
+            return detectBoundariesGeneric;
+    }
+}
+// =============================================================================
+// 切块主入口
+// =============================================================================
+/**
+ * 把单个文件内容切成 chunks。
+ *
+ * 策略：
+ *   1. 用语言对应的正则识别边界点
+ *   2. 在每两个边界之间组成 chunk（如果合并段超 MAX_CHUNK_CHARS，按 HARD_LINES 硬切）
+ *   3. 没有边界时按 HARD_LINES 硬切
+ *
+ * 行号统一 1-indexed inclusive。
+ */
+export function chunkFile(filePath, content) {
+    const lines = content.split(/\r?\n/);
+    if (lines.length === 0)
+        return [];
+    const ext = filePath.slice(filePath.lastIndexOf("."));
+    const detect = pickDetector(ext);
+    const boundaries = detect(lines);
+    // 把"文件起点"作为虚拟边界
+    const cuts = [{ line: 1, kind: "header", name: "" }, ...boundaries];
+    const chunks = [];
+    // 按相邻边界切片
+    for (let i = 0; i < cuts.length; i++) {
+        const start = cuts[i].line;
+        const end = i + 1 < cuts.length ? cuts[i + 1].line - 1 : lines.length;
+        if (end < start)
+            continue;
+        sliceAndPush(lines, filePath, start, end, cuts[i].kind, cuts[i].name, chunks);
+    }
+    return chunks;
+}
+function sliceAndPush(lines, _filePath, startLine, endLine, kind, name, out) {
+    // 把 [startLine..endLine] 视作一个 region，超长就硬切
+    let cursor = startLine;
+    while (cursor <= endLine) {
+        const lineCount = Math.min(endLine - cursor + 1, HARD_LINES);
+        let endCursor = cursor + lineCount - 1;
+        let body = lines.slice(cursor - 1, endCursor).join("\n");
+        // 字符上限再约束一次（超长则进一步缩 endCursor）
+        while (body.length > MAX_CHUNK_CHARS && endCursor > cursor) {
+            endCursor = cursor + Math.max(0, Math.floor((endCursor - cursor) * 0.8));
+            body = lines.slice(cursor - 1, endCursor).join("\n");
+        }
+        if (body.trim().length === 0) {
+            cursor = endCursor + 1;
+            continue;
+        }
+        out.push({
+            line_start: cursor,
+            line_end: endCursor,
+            symbol_kind: kind,
+            symbol_name: name,
+            content: body,
+        });
+        cursor = endCursor + 1;
+    }
+}
+// =============================================================================
+// 文件类型过滤（黑名单）
+// =============================================================================
+const SOURCE_EXTS = new Set([
+    ".java", ".kt", ".kts", ".aidl", ".hal",
+    ".c", ".cc", ".cpp", ".cxx", ".h", ".hh", ".hpp",
+    ".py", ".sh", ".bp", ".mk", ".rs", ".go",
+    ".xml", ".rc", ".te", ".sepolicy",
+    ".js", ".ts", ".jsx", ".tsx", ".vue",
+    ".md", ".txt",
+]);
+const SKIP_DIR_NAMES = new Set([
+    ".git", ".repo", "out", "build", ".gradle", "node_modules",
+    "__pycache__", ".idea", ".vscode",
+    "obj", "bin", "Debug", "Release", "target",
+]);
+/** 是否应该索引该文件路径（基于扩展名 + 大小）。 */
+export function shouldIndexFile(filePath, sizeBytes) {
+    if (sizeBytes > 5 * 1024 * 1024)
+        return false; // 单文件 5MB 上限
+    if (sizeBytes === 0)
+        return false;
+    const i = filePath.lastIndexOf(".");
+    if (i < 0)
+        return false;
+    const ext = filePath.slice(i).toLowerCase();
+    return SOURCE_EXTS.has(ext);
+}
+/** 是否应该跳过目录（按目录名）。 */
+export function shouldSkipDir(dirName) {
+    if (dirName.startsWith(".") && SKIP_DIR_NAMES.has(dirName))
+        return true;
+    return SKIP_DIR_NAMES.has(dirName);
+}

package/dist/aosp/embed-aosp.js

@@ -0,0 +1,56 @@
+/**
+ * AOSP chunks 的嵌入回写。
+ *
+ * 与三源 embed_pending 拆开：aosp 通常 chunk 数量级巨大（几十万），
+ * 单独控制 batch / progress 更合理。
+ */
+import { getDb, runInTransaction, vectorToBlob } from "../db.js";
+import { embedTexts } from "../embedder.js";
+import { invalidateAospIndex } from "./search.js";
+export async function embedAospPending(args = {}) {
+    const limit = Math.max(1, Math.min(5000, args.batch_size ?? 200));
+    const db = getDb();
+    const where = ["embedding IS NULL"];
+    const params = [];
+    if (args.platform) {
+        where.push("platform = ?");
+        params.push(args.platform.toUpperCase());
+    }
+    if (args.module) {
+        where.push("module = ?");
+        params.push(args.module.toLowerCase());
+    }
+    // 1. 总待嵌入数（用于进度报告）
+    const totalRow = db
+        .prepare(`SELECT COUNT(*) AS n FROM aosp_chunks WHERE ${where.join(" AND ")}`)
+        .get(...params);
+    const total = totalRow?.n ?? 0;
+    if (total === 0)
+        return { total_pending: 0, embedded: 0, remaining: 0 };
+    // 2. 拉本批
+    const rows = db
+        .prepare(`SELECT id, content FROM aosp_chunks WHERE ${where.join(" AND ")} ORDER BY id LIMIT ?`)
+        .all(...params, limit);
+    if (rows.length === 0)
+        return { total_pending: total, embedded: 0, remaining: total };
+    // 3. embed
+    const texts = rows.map((r) => r.content);
+    const vectors = await embedTexts(texts);
+    if (vectors.length !== rows.length) {
+        throw new Error(`embedTexts 返回 ${vectors.length}，期望 ${rows.length}`);
+    }
+    // 4. 回写
+    const ts = new Date().toISOString();
+    const stmt = db.prepare("UPDATE aosp_chunks SET embedding = ?, embedding_updated_at = ? WHERE id = ?");
+    runInTransaction(db, () => {
+        for (let i = 0; i < rows.length; i++) {
+            stmt.run(vectorToBlob(vectors[i]), ts, rows[i].id);
+        }
+    });
+    invalidateAospIndex();
+    return {
+        total_pending: total,
+        embedded: rows.length,
+        remaining: total - rows.length,
+    };
+}